# A Copy-Augmented Sequence-to-Sequence Architecture Gives Good Performance on Task-Oriented Dialogue

## TL;DR

This paper demonstrates that a simple copy-augmented sequence-to-sequence neural model can effectively handle task-oriented dialogue, outperforming complex models and matching state-of-the-art results without explicit intent or belief state modeling.

## Contribution

It introduces a copy-augmented sequence-to-sequence architecture that bypasses explicit state modeling, achieving competitive performance in task-oriented dialogue.

## Key findings

- Outperforms memory-augmented models by 7% in per-response accuracy
- Matches current state-of-the-art on DSTC2 dataset
- Shows effectiveness of simple seq2seq with copy mechanism

## Abstract

Task-oriented dialogue focuses on conversational agents that participate in user-initiated dialogues on domain-specific topics. In contrast to chatbots, which simply seek to sustain open-ended meaningful discourse, existing task-oriented agents usually explicitly model user intent and belief states. This paper examines bypassing such an explicit representation by depending on a latent neural embedding of state and learning selective attention to dialogue history together with copying to incorporate relevant prior context. We complement recent work by showing the effectiveness of simple sequence-to-sequence neural architectures with a copy mechanism. Our model outperforms more complex memory-augmented models by 7% in per-response generation and is on par with the current state-of-the-art on DSTC2.